IBM Global Services Relies on Fibre Channel

Posted on May 01, 1998

RssImageAltText

IBM Global Services Relies on Fibre Channel

Big Blue`s systems integration division installed a Fibre Channel network for maximum speed and reliability.

By Zachary Shess

Senior Editor

Anyone who`s dealt with leaky faucets knows how annoying it can be to fix them. Few of us, however, can relate to the larger task of repairing a leaky dam. When faced with just a slight data integrity problem, what would be a trickle to most IS teams is a potential breaking dam for IBM Global Services (IGS).

As the world`s largest systems integrator (with $19.3 billion in revenue last year), IGS provides networking services to some of the country`s largest companies, including Merrill Lynch, Monsanto, General Electric, and Sears Roebuck. IGS holds and manages hundreds of terabytes of mission-critical information for clients at their Research Triangle Park, NC, site alone, one of several IGS data centers around the world. Service Level Agreements require IGS to pay clients back if their data accessibility is compromised. A significant agreement lasting five to ten years can be worth as much as $10 billion. Bottom line: IGS can have no system failures.

Recently, this fastest growing division of IBM raised some eyebrows when they began implementing a Fibre Channel storage network to help provide data backup connections to several enterprise servers. IGS selected Fibre Channel over asynchronous transfer mode (ATM), Gigabit Ethernet, and IBM`s own Serial Storage Architecture (SSA) primarily because of Fibre Channel`s high-speed throughput and fault tolerance.

"IBM allows us to use whatever technology we need to solve a problem, so we build what`s best for the customer," says Ron Howell, an IGS network architect who, among other accomplishments, built Merrill Lynch`s worldwide email system.

IGS`s network is based on Ancor Communications` GigWorks Fibre Channel switches. Each evening, the switches move hundreds of terabytes of data from remote buildings within the Research Triangle Park site to one large primary backup center located about two miles away. Howell says that the advantages of Fibre Channel over other interface interconnects are unmatched speed, reliability, and efficiency, as well as lower cost per megabyte.

Easy Installation, Configuration

Howell says that installing Fibre Channel switches is easier than setting up, say, ATM switches. Installing an ATM switch requires setting the ATM ID number, followed by configuring and tuning the switch software, which can take half a day, according to Howell. With Fibre Channel, everything is already burned into ASICs, so once it`s installed, the only task left is setting the address.

Because of load aggregation, Howell believes that Fibre Channel networks are more reliable than other networks he has designed using Gigabit Ethernet or ATM. "If you have one pipe of Fibre Channel and you add a few others, it looks like one big bandwidth. So I can disable a line and the system doesn`t notice. With ATM, you have to break a line. And Gigabit Ethernet has to perform a spanning tree command, which takes 30 seconds to look for an alternate route."

How big a deal is 30 seconds? Based on the enormous amount of mission-critical data stored at the Research Triangle Park facility alone, Howell estimates that lost productivity from downtime costs more than $1 million per hour.

Testing Is Paramount

To create "bulletproof" systems, IGS extensively tests equipment it plans to implement. Network architects first set up a testing matrix, where they list performance requirements based on objectives determined by clients. Once this checklist is set up, the testing itself could be characterized more as getting components to break rather than observing how smoothly they work. "You want to test all the components because you`re only as strong as your weakest link," says Howell.

Fibre Channel switches have been utilized at IGS for about six months, only after more than two months of testing. During that period, Howell conducted load generations and simulated backups trying to generate frame losses. To his surprise, he did not have any frame losses during testing, and still has not had any in the six months of live implementation.

During the testing phase, Howell was assisted by Ancor, whose engineers helped tune the switches. The session included running Ancor`s NetPerf testing software, which took about 30 minutes to execute. NetPerf tells engineers the actual speed of the Fibre Channel pipe. Howell says the testing tool is valuable to gauge the network`s true efficiency. By mid-March, IGS was achieving 94% of Fibre Channel`s theoretical speed of 100MBps per loop. With ATM, Howell says he`s lucky if he can get 60% to 70% efficiency.

Once test results are finalized, Howell believes it is imperative to share information among the engineers. In fact, he estimates that nearly one-third of his job is sharing test results.

Howell believes it is essential to use a non-blocking switch, such as Ancor`s GigWorks. "A switch is a highway that lets you get from Point A to Point B. If you have 15 on-ramps, the data should be able to get in and out of the switch without blocking anyone else. If you have blocking, some of the highways will not be able to get in at the same time."

Howell is quick to point out that even if you`re not the size of an IGS, you still have to be accountable for the product you produce or service. When billions of dollars are at stake, that accountability is magnified even more. While few system integrators may be able to relate to the size and scope of IGS, their practice of extensive testing and their unwillingness to compromise their objectives can be a model for any integrator`s success.

Click here to enlarge image

At its Research Triangle Park facility alone, IGS holds and manages terabytes of mission-critical data for some of the nation`s largest companies, including Sears Roebuck and Merrill Lynch.

Click here to enlarge image

Each evening, Ancor`s Fibre Channel switches move data from remote buildings to one central backup center.


Comment and Contribute
(Maximum characters: 1200). You have
characters left.